OPEN EPHEMERIS
Back to blog
ChatGPT Gets the Planets Right Now. The Rising Sign Is Another Story.

November 13, 2026

·

Jackson

ChatGPT Gets the Planets Right Now. The Rising Sign Is Another Story.

We gave four AI chatbots the same 20 birth charts and checked every answer against NASA data. The planets are mostly right now. On births near a clock change, the Rising sign and houses still go wrong, with no warning.

aiastrologybirth-chartaccuracychatgptgeminigrok

More and more people ask an AI chatbot for their birth chart. So we checked how often the answer is right.

We gave ChatGPT, Gemini, Grok and DeepSeek the same 20 births, the way an ordinary person uses them: the free app, default settings, a fresh private chat each time, the same plain question. Then we checked every placement against NASA's data.

The short version:

  • The planets are mostly right now. ChatGPT and Grok calculate them instead of guessing.
  • The Rising sign and the houses are not. On births near a clock change or a time-zone change, ChatGPT got half the charts fully right and Grok two in ten.
  • None of the wrong charts said they might be wrong.

What "fully right" means

A birth chart has two parts.

  1. Where the planets were. The Sun, Moon and planets, each in a sign.
  2. The frame. The Rising sign (the sign coming up on the eastern horizon at the minute you were born) and the twelve houses that follow from it. The houses say which part of life each planet is about: work, home, relationships, self.

We counted a chart as fully right only if every planet and the Rising sign were within 1° of NASA's position, in the right sign, and every planet was in the right house.

Ordinary births: the planets are right now

On 10 ordinary births (1973 to 2004, cities with simple time-zone histories):

AppFully right
ChatGPT9 of 10
Grok7 of 10
Gemini3 of 10
DeepSeekdeclined every time

ChatGPT and Grok write and run a small program inside the chat, and the positions come from a real astronomy library. Their planets were within a small fraction of a degree. This is a real improvement. If you have read that AI gets 49 of 50 charts wrong, that is no longer true of ChatGPT.

Gemini also ran code, but its Moon was often several degrees off. DeepSeek said it could not calculate a chart and that a made-up one would be worse than none. That is honest, but it leaves you with nothing.

Harder births: the Rising sign breaks

Then we tried 10 births where the clock itself is the hard part. Nothing exotic, just days that real people are born on:

  • an hour before the clocks went back
  • the year Alaska changed time zones
  • Iran's half-hour time zone
  • New Zealand's 1970s summer-time trial
  • a town inside the Arctic Circle
AppFully right
ChatGPT5 of 10
Grok2 of 10
Gemini2 of 10

The planets mostly stayed right. The frame did not.

  • Los Angeles, October 1972, 12:31 a.m. ChatGPT put the Rising sign three signs away from where it belongs. Grok put it on the exact opposite side of the chart.
  • Hobart, March 1971. Grok put the Rising sign almost directly opposite the real one, and nine of ten planets landed in the wrong house.
  • Auckland, March 1976. ChatGPT had every planet right and five of them in the wrong house.

Each of these came back as a clean, confident chart. Nothing on the screen told the reader which part to doubt.

Why a small error is a big one

People rarely ask an AI about their chart on a quiet afternoon. They ask after a breakup, during a move, when a parent is ill, when a job ends. They want help understanding what is happening to them.

The Rising sign and houses are the foundation of that reading. Move the Rising sign and every house moves with it. A transit that is about your work gets read as being about your marriage. Advice built on it points the wrong way.

So "mostly right" is not good enough here. One wrong sign at the base makes the whole reading wrong, and it reaches someone at the moment they are most likely to act on it.

Why the chatbots get this part wrong

The planets are the easy half. Any good astronomy library has them.

The hard half is time. To place the Rising sign you need the exact moment of birth in universal time. That depends on what the clocks said in that town on that day: which daylight-saving rule applied, whether there was a war-time or trial rule, whether the town had changed zones. A chatbot writes fresh code for this in every chat. Sometimes it gets the history right. Sometimes it does not, and it cannot tell you which.

How to check your own chart

If you were born near a clock change, outside a big city, before about 1980, or far north or south, check your Rising sign before you build a reading on it.

  1. Find the time on your birth certificate, not a family memory of it.
  2. Get your chart from a source that uses a real historical time-zone database and shows the time offset it used.
  3. Compare the Rising sign and its degree with what your chatbot told you. If they differ, trust the calculated one.

The fix: let the AI talk and a real engine calculate

Chatbots are good at explaining meaning. They are not reliable calculators. The fix is to give them one.

Open Ephemeris is that calculator, built to plug into the chat you already use:

  • Positions from NASA JPL DE440, the same data used to navigate spacecraft, checked against NASA's own Horizons service.
  • A full historical time-zone database, so births near clock changes and zone changes get the right moment, and the chart shows the offset it used.
  • The same answer every time. The chatbot doesn't write new code for each chat; it asks one tested engine.
  • Proven in the same kind of test. When we connected Open Ephemeris to Claude, 209 of 210 charts were fully right, up from 1 of 210 on its own. See that test.
  • More than Western astrology. Vedic charts, Human Design, Chinese BaZi, transits and compatibility come from the same engine.

Start here:

It's free to start, with no card needed. The chatbot still does the talking, but the chart underneath is real.

What this test can and cannot say

  • Small numbers. 20 births per app, one try each, on one day. A result of 5 in 10 could be anywhere from about 2 in 10 to 8 in 10.
  • The harder births were chosen to be hard. Their failure rate is not a rate for every birth.
  • Kimi was not tested, and we did not try paid tiers.
  • We make Open Ephemeris. This is our test, not an independent one. The answer key is the same NASA-checked set used in our first test.